Reflection Pretraining Enables Token-Level Self-Correction in Biological Sequence Models

Chain-of-Thought (CoT) prompting has significantly advanced task-solving capabilities in natural language processing with large language models. Unlike standard prompting, CoT encourages the model to generate intermediate reasoning steps, non-answer tokens, that help guide the model toward more accurate final outputs. These intermediate steps enable more complex reasoning processes such as error correction, memory management, future planning, and self-reflection. However, applying CoT to non-natural language domains, such as protein and RNA language models, is not yet possible, primarily due to the limited expressiveness of their token spaces (e.g., amino acid tokens). In this work, we propose and define the concept of language expressiveness: the ability of a given language, using its tokens and grammar, to encode information. We show that the limited expressiveness of protein language severely restricts the applicability of CoT-style reasoning. To overcome this, we introduce reflection pretraining, for the first time in a biological sequence model, which enables the model to engage in intermediate reasoning through the generation of auxiliary "thinking tokens" beyond simple answer tokens. Theoretically, we demonstrate that our augmented token set significantly enhances biological language expressiveness, thereby improving the overall reasoning capacity of the model. Experimentally, our pretraining approach teaches protein models to self-correct and leads to substantial performance gains compared to standard pretraining.

Distilling the Essence: Efficient Reasoning Distillation via Sequence Truncation

Distilling the reasoning capabilities from a large language model (LLM) to a smaller student model often involves training on substantial amounts of reasoning data. However, distillation over lengthy sequences with prompt (P), chain-of-thought (CoT), and answer (A) segments makes the process computationally expensive. In this work, we investigate how the allocation of supervision across different segments (P, CoT, A) affects student performance. Our analysis shows that selective knowledge distillation over only the CoT tokens can be effective when the prompt and answer information is encompassed by it. Building on this insight, we establish a truncation protocol to quantify computation-quality tradeoffs as a function of sequence length. We observe that training on only the first $50\%$ of tokens of every training sequence can retain, on average, $\approx94\%$ of full-sequence performance on math benchmarks while reducing training time, memory usage, and FLOPs by about $50\%$ each. These findings suggest that reasoning distillation benefits from prioritizing early reasoning tokens and provides a simple lever for computation-quality tradeoffs. Codes are available at https://github.com/weiruichen01/distilling-the-essence.

AraHealthQA 2025: The First Shared Task on Arabic Health Question Answering

We introduce {AraHealthQA 2025}, the {Comprehensive Arabic Health Question Answering Shared Task}, held in conjunction with {ArabicNLP 2025} (co-located with EMNLP 2025). This shared task addresses the paucity of high-quality Arabic medical QA resources by offering two complementary tracks: {MentalQA}, focusing on Arabic mental health Q\&A (e.g., anxiety, depression, stigma reduction), and {MedArabiQ}, covering broader medical domains such as internal medicine, pediatrics, and clinical decision making. Each track comprises multiple subtasks, evaluation datasets, and standardized metrics, facilitating fair benchmarking. The task was structured to promote modeling under realistic, multilingual, and culturally nuanced healthcare contexts. We outline the dataset creation, task design and evaluation framework, participation statistics, baseline systems, and summarize the overall outcomes. We conclude with reflections on the performance trends observed and prospects for future iterations in Arabic health QA.

Pearl: A Multimodal Culturally-Aware Arabic Instruction Dataset

Mainstream large vision-language models (LVLMs) inherently encode cultural biases, highlighting the need for diverse multimodal datasets. To address this gap, we introduce Pearl, a large-scale Arabic multimodal dataset and benchmark explicitly designed for cultural understanding. Constructed through advanced agentic workflows and extensive human-in-the-loop annotations by 45 annotators from across the Arab world, Pearl comprises over K multimodal examples spanning ten culturally significant domains covering all Arab countries. We further provide two robust evaluation benchmarks Pearl and Pearl-Lite along with a specialized subset Pearl-X explicitly developed to assess nuanced cultural variations. Comprehensive evaluations on state-of-the-art open and proprietary LVLMs demonstrate that reasoning-centric instruction alignment substantially improves models' cultural grounding compared to conventional scaling methods. Pearl establishes a foundational resource for advancing culturally-informed multimodal modeling research. All datasets and benchmarks are publicly available.

Voice of a Continent: Mapping Africa's Speech Technology Frontier

Africa's rich linguistic diversity remains significantly underrepresented in speech technologies, creating barriers to digital inclusion. To alleviate this challenge, we systematically map the continent's speech space of datasets and technologies, leading to a new comprehensive benchmark SimbaBench for downstream African speech tasks. Using SimbaBench, we introduce the Simba family of models, achieving state-of-the-art performance across multiple African languages and speech tasks. Our benchmark analysis reveals critical patterns in resource availability, while our model evaluation demonstrates how dataset quality, domain diversity, and language family relationships influence performance across languages. Our work highlights the need for expanded speech technology resources that better reflect Africa's linguistic diversity and provides a solid foundation for future research and development efforts toward more inclusive speech technologies.

NileChat: Towards Linguistically Diverse and Culturally Aware LLMs for Local Communities

Enhancing the linguistic capabilities of Large Language Models (LLMs) to include low-resource languages is a critical research area. Current research directions predominantly rely on synthetic data generated by translating English corpora, which, while demonstrating promising linguistic understanding and translation abilities, often results in models aligned with source language culture. These models frequently fail to represent the cultural heritage and values of local communities. This work proposes a methodology to create both synthetic and retrieval-based pre-training data tailored to a specific community, considering its (i) language, (ii) cultural heritage, and (iii) cultural values. We demonstrate our methodology using Egyptian and Moroccan dialects as testbeds, chosen for their linguistic and cultural richness and current underrepresentation in LLMs. As a proof-of-concept, we develop NileChat, a 3B parameter LLM adapted for Egyptian and Moroccan communities, incorporating their language, cultural heritage, and values. Our results on various understanding, translation, and cultural and values alignment benchmarks show that NileChat outperforms existing Arabic-aware LLMs of similar size and performs on par with larger models. We share our methods, data, and models with the community to promote the inclusion and coverage of more diverse communities in LLM development.

Where Are We? Evaluating LLM Performance on African Languages

Africa's rich linguistic heritage remains underrepresented in NLP, largely due to historical policies that favor foreign languages and create significant data inequities. In this paper, we integrate theoretical insights on Africa's language landscape with an empirical evaluation using Sahara - a comprehensive benchmark curated from large-scale, publicly accessible datasets capturing the continent's linguistic diversity. By systematically assessing the performance of leading large language models (LLMs) on Sahara, we demonstrate how policy-induced data variations directly impact model effectiveness across African languages. Our findings reveal that while a few languages perform reasonably well, many Indigenous languages remain marginalized due to sparse data. Leveraging these insights, we offer actionable recommendations for policy reforms and inclusive data practices. Overall, our work underscores the urgent need for a dual approach - combining theoretical understanding with empirical evaluation - to foster linguistic diversity in AI for African communities.

Cross-Modal Consistency in Multimodal Large Language Models

Recent developments in multimodal methodologies have marked the beginning of an exciting era for models adept at processing diverse data types, encompassing text, audio, and visual content. Models like GPT-4V, which merge computer vision with advanced language processing, exhibit extraordinary proficiency in handling intricate tasks that require a simultaneous understanding of both textual and visual information. Prior research efforts have meticulously evaluated the efficacy of these Vision Large Language Models (VLLMs) in various domains, including object detection, image captioning, and other related fields. However, existing analyses have often suffered from limitations, primarily centering on the isolated evaluation of each modality's performance while neglecting to explore their intricate cross-modal interactions. Specifically, the question of whether these models achieve the same level of accuracy when confronted with identical task instances across different modalities remains unanswered. In this study, we take the initiative to delve into the interaction and comparison among these modalities of interest by introducing a novel concept termed cross-modal consistency. Furthermore, we propose a quantitative evaluation framework founded on this concept. Our experimental findings, drawn from a curated collection of parallel vision-language datasets developed by us, unveil a pronounced inconsistency between the vision and language modalities within GPT-4V, despite its portrayal as a unified multimodal model. Our research yields insights into the appropriate utilization of such models and hints at potential avenues for enhancing their design.

Swan and ArabicMTEB: Dialect-Aware, Arabic-Centric, Cross-Lingual, and Cross-Cultural Embedding Models and Benchmarks

We introduce Swan, a family of embedding models centred around the Arabic language, addressing both small-scale and large-scale use cases. Swan includes two variants: Swan-Small, based on ARBERTv2, and Swan-Large, built on ArMistral, a pretrained Arabic large language model. To evaluate these models, we propose ArabicMTEB, a comprehensive benchmark suite that assesses cross-lingual, multi-dialectal, multi-domain, and multi-cultural Arabic text embedding performance, covering eight diverse tasks and spanning 94 datasets. Swan-Large achieves state-of-the-art results, outperforming Multilingual-E5-large in most Arabic tasks, while the Swan-Small consistently surpasses Multilingual-E5 base. Our extensive evaluations demonstrate that Swan models are both dialectally and culturally aware, excelling across various Arabic domains while offering significant monetary efficiency. This work significantly advances the field of Arabic language modelling and provides valuable resources for future research and applications in Arabic natural language processing. Our models and benchmark will be made publicly accessible for research.

Desert Camels and Oil Sheikhs: Arab-Centric Red Teaming of Frontier LLMs

Large language models (LLMs) are widely used but raise ethical concerns due to embedded social biases. This study examines LLM biases against Arabs versus Westerners across eight domains, including women's rights, terrorism, and anti-Semitism and assesses model resistance to perpetuating these biases. To this end, we create two datasets: one to evaluate LLM bias toward Arabs versus Westerners and another to test model safety against prompts that exaggerate negative traits ("jailbreaks"). We evaluate six LLMs -- GPT-4, GPT-4o, LlaMA 3.1 (8B & 405B), Mistral 7B, and Claude 3.5 Sonnet. We find 79% of cases displaying negative biases toward Arabs, with LlaMA 3.1-405B being the most biased. Our jailbreak tests reveal GPT-4o as the most vulnerable, despite being an optimized version, followed by LlaMA 3.1-8B and Mistral 7B. All LLMs except Claude exhibit attack success rates above 87% in three categories. We also find Claude 3.5 Sonnet the safest, but it still displays biases in seven of eight categories. Despite being an optimized version of GPT4, We find GPT-4o to be more prone to biases and jailbreaks, suggesting optimization flaws. Our findings underscore the pressing need for more robust bias mitigation strategies and strengthened security measures in LLMs.